If extraction is what you want, you don't need to explicitly use RecursiveParserWrapper if you use AutodetectParser. Autodetect parser does it automatically for you. It sets RecursiveParserWrapper and by default sets ParsingEmbeddedDocumentExtractor as EmbeddedDocumentExteractor by default.
This extractor delegates parsing to the same instance of AutodetectParser. That is the embedded images also get parsed by the same AutodetectParser you passed the parent documents to. This parser will perform ocr extraction if any parser is in the registry supports ocr media types (eg: image/ocr-png, image/ocr-jpeg).
TesseractOCRParser will only register itself with ocr support types if tesseract is present in the host machine or the parser is ignored as it returns an empty set for supported types